DISK WINDS, JETS, AND OUTFLOWS:THEORETICAL AND COMPUTATIONAL FOUNDATIONS

Ralph E. Pudritz1, Rachid Ouyed2, Christian Fendt3, and Axel Brandenburg41Dept. of Physics & Astronomy, McMaster University, Hamilton, ON L8S 4M1, Canada2 Dept. of Physics & Astronomy, University of Calgary, Calgary, AB T2N 1N4, Canada3 Max Planck Institute for Astronomy, Königstuhl 17, D-69117 Heidelberg, Germany

4 Nordita, Blegdamsvej 17, DK-2100 Copenhagen Ø, Denmark

We review advances in the theoretical and computational studies of disk winds, jetsand outflows including: the connection between accretion and jets, the launch of jets frommagnetized disks, the coupled evolution of jets and disks, the interaction of magnetized youngstellar objects with their surrounding disks and the relevance to outflows, and finally, the linkbetween jet formation and gravitational collapse. We also address the important predictions ofthe theory on jet kinematics, collimation, and rotation, that have recently been confirmed byhigh spatial and spectral resolution HST observations. Finally, we show that disk winds have auniversal character that can account for jets and outflows during the formation of massive starsas well as brown dwarfs.

1. INTRODUCTION

The close association of jets and outflows with protostel-lar accretion disks is one of the hallmarks of the accretionpicture of low mass star formation. The most energetic out-flow phase occurs during gravitational collapse of a molec-ular cloud core - the so-called Class 0 phase - when muchof its envelope is still raining down onto the forming proto-stellar disk and the disk accretion rate is high. Later, in theT-Tauri star (TTS) stage, when most of the original core hasbeen accreted and the young stellar object (YSO) is beingfed by lower accretion rates through the surrounding Kep-lerian accretion disk, the high-speed jet becomes opticallyvisible. When the disk disappears in the weak-lined TTS(WTTS) phase, the jet goes with it.

The most comprehensive theoretical picture that we havefor these phenomena is that jets are highly collimated, hy-dromagnetic disk winds whose torques efficiently extractdisk angular momentum and gravitational potential energy.Jets also sweep up ambient molecular gas and drive largescale molecular outflows. A disk wind was first suggestedas the origin of jets from accretion disks around black holesin the seminal paper byBlandford and Payne(1982, BP82),and was soon proposed as the mechanism for protostellarjets (Pudritz and Norman, 1983, 1986).

Several major observational breakthroughs have takenplace in the study of jets and outflows since PPIV (held in1998). Direct, high resolution spectro-imaging and adap-tive optics methods discovered the rotation of protostellarjets (e.g.Bacciotti et al., 2003). These observations also re-vealed that jets have an onion-like, velocity structure (withthe highest speeds being closest to the outflow axis). Thisbody of work provides strong support for the idea that jetsoriginate as centrifugally driven MHD winds from extendedregions of their surrounding disks (see chapter by Ray et

al.). Recently, outflows and disks have also been discoveredaround massive stars (see chapter by Arce et al.) as well asbrown dwarfs (eg.Bourke et al. 2005) implying that themechanism is of importance across the entire stellar massspectrum.

Major advances in the theoretical modeling of these sys-tems have also occurred, due primarily to a variety of MHDcomputational studies. Simulations can now resolve theglobal evolution of disks and outflows, track the interactionof disks with central magnetized stars, and even follow thegeneration of outflows during gravitational collapse. Thisbody of work shows that jets and disks are closely coupledand that the basic dynamics of jets scales to YSOs of allmasses.

Our review examines the theory of the central engine ofjets and its exploration through the use of computer simu-lations. We focus mainly on developments since PPIV andrefer to the review byKönigl and Pudritz(2000, KP00) for adiscussion of the earlier literature, as well asPudritz(2003)andHeyvaerts(2003) for more technical background. Wefirst discuss the basic theory of disk winds and their kine-matics (§2). We review computational studies of jets fromaccretion disks treated as boundary conditions (§3), as wellas global simulations including the disk (§4). We then ex-amine the innermost regions of the disk where the stellarmagnetosphere interacts with the disk, as well as the sur-face of the star that may drive an accretion-powered outflow(§5). Finally, we discuss how outflows are generated duringthe early stages of the gravitational collapse (§6).

2. THEORY OF DISK WINDS

An important insight into the nature of the engine forjets can be gleaned from the observed ratio of the momen-tum transport rate (or thrust) carried by the CO molecular

1

outflow to the thrust that can be provided by the bolomet-ric luminosity of the central star (e.g.Cabrit and Bertout,1992);

Foutflow/Frad = 250(Lbol/103L�)

−0.3, (1)

This relation has been confirmed and extended by the analy-sis of data from over 390 outflows, ranging over six decadesup to106L� in stellar luminosity (Wu et al., 2004). It sug-gests that jets from both low and high mass systems areprobably driven by a single, non-radiative, mechanism.

Jets are observed to have a variety of structures andtime-dependent behaviour – from internal shocks and mov-ing knots to systems of bow shocks that suggest long-time episodic outbursts. They show wiggles and oftenhave cork-screw like structure, suggesting the presence ei-ther of jet precession, the operation of non-axisymmetrickink modes, or both. Given the highly nonlinear be-haviour of the force balance equation for jets (the so-calledGrad-Shafranov equation), theoretical work has focused ontractable and idealized time-independent, and axisymmetricor self-similar models (e.g. BP82) of various kinds.

2.1. Conservation Laws and Jet Kinematics

Conservation laws are the gold standard in physics, andplay a significant role in understanding astrophysical jets.This is because whatever the details (e.g. the asymptotics,the crossing of critical points, the way that matter is loadedonto field lines within the disks, etc.), conservation lawsstrongly constrain the flux of mass, angular momentum, andenergy. What cannot be constrained by these laws will de-pend on the general physics of the disks such as on howmatter is loaded onto field lines (§4).

Jet dynamics can be described by the time-dependent,equations of ideal MHD. The evolution of a magnetized,rotating system that is threaded by a large-scale fieldB in-volves: the continuity equation for a conducting gas of den-sity ρ moving at velocityv (which includes turbulence); theequation of motion for the gas which undergoes pressure(p), gravitational (with potentialΦ), and Lorentz forces;the induction equation for the evolution of the magneticfield in the moving gas where the current density isj =(c/4π)∇× B; the energy equation, wheree the internal en-ergy per unit mass; and the absence of magnetic monopoles;which are written, respectively:

∂ρ

∂t+ ∇.(ρv) = 0 (2)

ρ

(

∂v

∂t+ (v.∇)v

)

+ ∇p + ρ∇Φ −j × B

c= 0 (3)

∂B

∂t−∇× (v × B) = 0 (4)

ρ

(

∂e

∂t+ (v.∇)e

)

+ p(∇.v) = 0 (5)

∇.B = 0 (6)

We specialize to the restricted case of stationary, as wellas 2D (axisymmetric) flows, from which the conservation

laws follow. It is useful to decompose vector quantitiesinto poloidal and toroidal components (e.g. magnetic fieldB = Bp+Bφêφ). In axisymmetric conditions, the poloidalfieldBp can be derived from a single scalar potentiala(r, z)whose individual values,a = const, define the surfaces ofconstant magnetic flux in the outflow and can be specifiedat the surface of the disk (e.g.,Pelletier and Pudritz, 1992;PP92).

Conservation of mass and magnetic fluxalong a fieldline can be combined into a single functionk that is calledthe “mass load” of the wind which is a constant along amagnetic field line;

ρvp = kBp. (7)

This function represents the mass load per unit time, perunit magnetic flux of the wind. For axisymmetric flows, itsvalue is preserved on each ring of field lines emanating fromthe accretion disk. Its value on each field line is determinedby physical conditions - including dissipative processes -near the disk surface. It may be more revealingly written as

k(a) =ρvpBp

=dṀwdΨ

, (8)

wheredṀw is the mass flow rate through an annulus ofcross-sectional areadA through the wind anddΨ is theamount of poloidal magnetic flux threading through thissame annulus. The mass load profile is a function of thefootpoint radiusr0 of the wind on the disk.

The toroidal field in rotating flows derives from the in-duction equation;

Bφ =ρ

k(vφ − Ω0r), (9)

whereΩ0 is the angular velocity of the disk at the mid-plane. This result shows that the strength of the toroidalfield in the jet depends on the mass loading as well as the jetdensity. Denser winds should have stronger toroidal fields.We note however, that the density does itself depend on thevalue ofk. Equation (8) also suggests that at higher massloads, one has lower toroidal field strengths. This can bereconciled however, since it can be shown from the conser-vation laws (see below) that the value ofk is related to thedensity of the outflow at the Alfvén point on a field line;k = (ρA/4π)

1/2 (eg. PP92). Thus, higher mass loads cor-respond to denser winds and when this is substituted intoequation (8), we see that this also implies stronger toroidalfields. We show later that jet collimation depends on hoopstress through the toroidal field and thus the mass load musthave a very important effect on jet collimation (§3.2).

Conservation of angular momentumalong each fieldline leads to the conserved angular momentum per unitmass;

l(a) = rvφ −rBφ4πk

= const. (10)

The form forl reveals that the total angular momentum iscarried by both the rotating gas (first term) as well by the

2

twisted field (second term), the relative proportion beingdetermined by the mass load.

The value ofl(a) that is transported along each field lineis fixed by the position of the Alfvén point in the flow, wherethe poloidal flow speed reaches the Alfvén speed for thefirst time (mA = 1). It is easy to show that the value of thespecific angular momentum is;

l(a) = Ω0r2A = (rA/r0)

2l0. (11)

wherel0 = vK,0r0 = Ω0r20 is the specific angular momen-tum of a Keplerian disk. For a field line starting at a pointr0on the rotor (disk in our case), the Alfvén radius isrA(r0)and constitutes a lever arm for the flow. The result showsthat the angular momentum per unit mass that is being ex-tracted from the disk by the outflow is a factor of(rA/r0)2

greater than it is for gas in the disk. For typical lever arms,one particle in the outflow can carry the angular momentumof ten of its fellows left behind in the disk.

Conservation of energyalong a field line is expressedas a generalized version of Bernoulli’s theorem (this may bederived by taking the dot product of the equation of motionwith Bp). Thus, there is a specific energye(a) that is a con-stant along field lines, which may be found in many papers(e.g. BP82 and PP92). Since the terminal speedvp = v∞ ofthe disk wind is much greater than its rotational speed, andfor cold flows, the pressure may also be ignored, one findsthe result:

v∞ ' 21/2Ω0rA = (rA/r0)vesc,0. (12)

There are three important consequences for jet kinemat-ics here; (i) that the terminal speed exceeds the local escapespeed from its launch point on the disk by the lever arm ra-tio; (ii) the terminal speed scales with the Kepler speed as afunction of radius, so that the flow will have an onion-likelayering of velocities, the largest inside, and the smallest onthe larger scales, as seen in the observations; and (iii) thatthe terminal speed depends on the depth of the local grav-itational well at the footpoint of the flow – implying that itis essentially scalable to flows from disks around YSOs ofany mass and therefore universal.

Another useful form of the conservation laws is the com-bination of energy and angular momentum conservationto produce a new constant along a field line (e.g. PP92);j(a) ≡ e(a) − Ω0l(a). This expression has been used (An-derson et al., 2003) to deduce the rotation rate of the launchregion on the Kepler disk, where the observed jet rotationspeed isvφ,∞ at a radiusr∞ and which is moving in thepoloidal direction with a jet speed ofvp,∞. Evaluatingj fora cold jet at infinity and noting that its value (calculated atthe foot point) isj(a0) = −(3/2)v2K,0, one solves for theKepler rotation at the point on the disk where this flow waslaunched:

Ω0 ' v2p,∞/ (2vφ,∞r∞) . (13)

When applied to the observed rotation of the Large Veloc-ity Component (LVC) of the jet DG Tau (Bacciotti et al.,

2002), this yields a range of disk radii for the observed ro-tating material in the range of disk radii, 0.3–4 AU, and themagnetic lever arm isrA/r0 ' 1.8–2.6.

2.2. Angular Momentum Extraction

How much angular momentum can such a wind extractfrom the disk? The angular momentum equation for theaccretion disk undergoing an external magnetic torque maybe written:

Ṁad(r0v0)

dr0= −r20BφBz |r0,H , (14)

where we have ignored transport by MRI turbulence or spi-ral waves. By using the relation between poloidal field andoutflow on the one hand, as well as the link between thetoroidal field and rotation of the disk on the other, the angu-lar momentum equation for the disk yields one of the mostprofound scaling relations in disk wind theory – namely –the link between disk accretion and mass outflow rate (seeKP00, PP92 for details):

Ṁa ' (rA/r0)2Ṁw. (15)

The observationally well known result that in many sys-tems,Ṁw/Ṁa ' 0.1 is a consequence of the fact that leverarms are often found in numerical and theoretical work tobe rA/r0 ' 3 – the observations of DG Tau being a per-fect example. Finally, we note that the angular momentumthat is observed to be carried by these rotating flows (e.g.DG Tau) is a consistent fraction of the excess disk angularmomentum – from 60–100% (e.g.,Bacciotti, 2004), whichis consistent with the high extraction efficiency discussedhere.

2.3. Jet Power and Universality

These results can be directly connected to the observa-tions of momentum and energy transport in the molecularoutflows. Consider the total mechanical energy that is car-ried by the jet, which may be written as (eg.Pudritz2003);

Ljet =12

∫ rj

ri

dṀwv2∞ '

GM∗Ṁa2ri

(

1 −r2ir2j

)

' 12Lacc.

(16)This explains the observations of Class 0 outflows whereinLw/Lbol ' 1/2, since the main luminosity of the centralsource at this time is due to accretion and not nuclear reac-tions. Later, this ratio decreases since the accretion lumi-nosity (which scales with the mechanical power of the jet)becomes relatively small compared to the bolometric lumi-nosity of the star as it nears the ZAMS.

This result states that the wind luminosity taps the gravi-tational energy release through accretion in the gravitationalpotential of the central object – and is a direct consequenceof Bernoulli’s theorem. This, and the previous results, im-ply that jets may be produced in any accreting system. Thelowest mass outflow that has yet been observed corresponds

3

to a proto-brown dwarf of luminosity' 0.09L�, a stel-lar mass of only20 − 45MJup, and a very low mass disk< 10−4M� (Bourke et al.2005).

It should be possible, therefore, for lower luminosity jetsto be launched from the disks around Jovian mass plan-ets. Recent hydrodynamical simulations of circumstellaraccretion disks containing and building up an orbiting pro-toplanetary core have numerically proven the existence ofa circum-planetary sub-disk in almost Keplerian rotationclose to the planet (Kley et al., 2001). The accretion rateof these sub-disks is abouṫMcp = 6 × 10−5 MJup yr−1

and is confirmed by many independent simulations. Withthat, the circum-planetary disk temperature may reach val-ues up to2000 K indicating a sufficient degree of ionizationfor matter-field coupling and would also allow for strongequipartition field strength (Fendt, 2003).

The possibility of a planetary scale MHD outflow, sim-ilar to the larger scale YSO disk winds, is indeed quitelikely because: (i) the numerically established existenceofcircum-planetary disks is a natural feature of the formationof massive planets; and (ii) the feasibility of a large-scalemagnetic field in the protoplanetary environment (Quillenand Trilling, 1998;Fendt, 2003). One may show, moreover,that the outflow velocity is of the order of the escape speedfor the protoplanet, at about60 km s−1 (Fendt, 2003).

On very general grounds, disk winds are likely to beactive during massive star formation (eg.Königl 1999).Such outflows may already start during the early collapsephase when the central YSO still has only a fraction of asolar mass (e.g.Pudritz and Banerjee, 2005). Such earlyoutflows may actually enhance the formation of massivestars via disk accretion by punching a hole in the infallingenvelop and releasing the building radiation pressure (e.g.Krumholz et al., 2005).

2.4. Jet Collimation: Wide-Angle vs. Highly Colli-mated Flows

In the standard picture of hydromagnetic winds, collima-tion of an outflow occurs because of the increasing toroidalmagnetic field in the flow resulting from the inertia of thegas. Beyond the Alfvén surface, equation (8) shows thatthe ratio of the toroidal field to the poloidal field in thejet is of the orderBφ/Bp ' r/rA � 1, so that the fieldbecomes highly toroidal. In this situation, collimation isachieved by the tension force associated with the toroidalfield which leads to a radially inwards directed componentof the Lorentz force (or “z-pinch”);FLorentz,r ' jzBφ. Thestability of such systems is examined in the next section.

In Heyvaerts and Norman(1989) it was shown that twotypes of solution are possible depending upon the asymp-totic behaviour of the total current intensity in the jet;

I = 2π

∫ r

0

jz(r′, z′)dr′ = (c/2)rBφ. (17)

In the limit that I → 0 as r → ∞, the field lines areparaboloids which fill space. On the other hand, if the cur-

rent is finite in this limit, then the flow is collimated to cylin-ders. The collimation of a jet therefore depends upon itscurrent distribution – and hence on the radial distributionofits toroidal field – which, as we saw earlier, depends on themass load. Mass loading therefore must play a very impor-tant role in controlling jet collimation.

It can be shown (Pudritz et al., 2006; PRO) that for apower law distribution of the magnetic field in the disk,Bz(r0, 0) ∝ r

µ−10 , with an injection speed at the base of

a (polytropic) corona that scales as the Kepler speed, thatthe mass load takes the formk ∝ r−1−µo . In this regime,the current takes the formI(r, z) ∝ r−µ−(1/2)o . Thus,the current goes to zero for models withµ < −1/2, andthat these therefore must be wide angle flows. For modelswith µ > −1/2 however, the current diverges, and the flowshould collimate to cylinders.

These results predict that jets should show different de-grees of collimation depending on how they are mass loaded(PRO). As an example, neither the highly centrally con-centrated, magnetic field lines associated with the initialsplit-monopole magnetic configuration used in simulationsby Romanova et al.(1997), nor the (similar field structureinvoked in the X-wind (reviewShu et al., 2000) should be-come collimated in this picture. On the other hand, less cen-trally (radially) concentrated magnetic configurations suchas the potential configuration ofOuyed and Pudritz(1997a,OPI) and BP82 should collimate to cylinders. We explorethis in §3.2 .

This result also explains the range of collimation that isobserved for molecular outflows. Models for observed out-flows fall into two general categories: the jet-driven bowshock picture, and a wind-driven shell picture in which themolecular gas is driven by an underlying wide-angle windcomponent such as given by the X-wind (e.g. reviewCabritet al., 1997). A survey of molecular outflows byLee et al.(2000) found that both mechanisms are needed in order toexplain the full set of systems observed.

Finally, we note that apart from these general theoremson collimation,Spruit et al.(1997) has proposed that a suffi-ciently strong poloidal field that is external to the flow couldalso force its collimation. According to this criterion, sucha poloidal field could collimate jets provided that the fieldstrength decreases no slower thanBp ∼ r−µ with µ ≤ 1.3.

3. NUMERICAL SIMULATIONS – DISKS AS JETENGINES

Computational approaches are necessary if we are toopen up and explore the vast spaces of solutions to thehighly nonlinear jet problem. The first simulations of non-steady MHD jets from accretion disks were performed byUchida and Shibata(1985) andShibata and Uchida(1986).These early simulations were based on initial states thatwere violently out of equilibrium and demonstrated the roleof magnetic fields in launching and accelerating jets to ve-locities of order of the Keplerian velocity of the disk.

The published simulations of non-radiative ideal MHD

4

YSO jets differ in their assumed initial conditions, such asthe magnetic field distribution on the disk, the conditions inthe plasma above the disk surfaces, the state of the initialdisk corona, and the handling of the gravity of the centralstar. Nevertheless, they share common goals: to establishand verify the four important stages in jet evolution, namely,(i) ejection; (ii) acceleration; (iii) collimation; and (iv) sta-bility. In the following, we describe a basic physical ap-proach for setting up clean numerical simulations and howthis leads to advances in our understanding of disk windsand outflows.

3.1. 2-Dimensional Simulations

The simulations we discuss here have been studied ingreater detail inOuyed and Pudritz(1997a; OPI);Ouyedand Pudritz (1997b; OPII);Ouyed et al.(1997; OPS);Ouyed and Pudritz(1999; OPIII); and PRO. We use thesefor pedagogical purposes in this review. To see anima-tions of the simulations presented here as well as those per-formed by other authors, the interested reader is directedto the “animations” link at “http://www.capca.ucalgary.ca”(hereafter CAPCA). These simulations were run using theZEUS (2-D, 3-D and MP) code which is arguably the bestdocumented and utilized MHD code in the literature (Stoneand Norman, 1992, 1994). It is an explicit, finite differencecode that runs on a staggered grid. The equations generallysolved are those of ideal MHD listed in§1, equations (1)–(5).

The evolution of the magnetic field (induction equa-tion (3) above) is followed by the method of constrainedtransport. In this approach, if∇.B = 0 holds for the initialmagnetic configuration, then it remains so for all later timesto machine accuracy. The obvious way of securing this con-dition is to use an initial vector potentialA(r, z, t = 0) thatdescribes the desired initial magnetic field at every point inthe computational domain.

To ensure a stable initial state that allows for a tractablesimulation, three simple rules are useful (OPS and OPI):(i) use a corona that in hydrostatic balance with the centralobject (to ensure a perfectly stable hydrostatic equilibriumof the corona, the point-mass gravitational potential shouldbe relocated to zone centers - see§4.2 in OP for details);(ii) put the disk in pressure balance with the corona aboveit, and (iii) use a force-free magnetic field configuration tothread the disk and corona. The initial magnetic configura-tions should be chosen so that no Lorentz force is exertedon the initial (non-rotating) hydrostatic corona describedabove; more specifically, it is wise to use initial force-freeor current-free configurations in the computational domain.

The BP82 self-similar solution for jets relies on the useof a simple polytropic equation of state, with an indexγ = 5/3. Their solution is possible because it eliminatesthe many complications that arise from the the energy equa-tion, while preserving the essence of the outflow problemThis Ansatz corresponds to situations where the combinedeffects of heating and cooling simulates a tendency toward a

locally constant entropy. This choice also greatly simplifiesthe numerical setup and allows one to test the code againstanalytic solutions.

Another key simplification in this approach is to exam-ine the physics of the outflow for fixed physical conditionsin the accretion disk. Thus, the accretion disk at the baseof the corona – and in pressure balance with the overlyingatmosphere – is given a density profile that it maintains atall times in the simulation, since the disk boundary condi-tions are applied to the “ghost zones” and are not part ofthe computational domain. In part, this simplification maybe justified by the fact that typically, accretion discs willevolve on longer time scales than their associated jets.

The hydrostatic state one arrives at has a simple ana-lytic solution which was used as the initial state for all ofour simulations and was adopted and further developed byseveral groups. These studies include e.g. low plasma-βmonopole-like field distributions (Romanova et al., 1997);a magneto-gravitational switch mechanism (Meier et al.,1997); the disk accretion-ejection process (Kudoh et al.,1998); the interrelation between the grid shape and the de-gree of flow collimation (Ustyugova et al., 1998); a self-adjusting disk magnetic field inclination (Krasnopolsky etal., 1999, 2003); the interrelation between the jet’s turbulentmagnetic diffusivity and collimation (Fendt andČemeljíc,2002); a varation of the disk rotation profile (Vitorino etal., 2002); or dynamo maintained disk magnetic fields (vonRekowski and Brandenburg, 2004). More elaborate setupswherein the initial magnetic field configuration originateson the surface of a star and connects with the disk have alsobeen examined (§5 – e.g.,Hayashi et al., 1996;Goodsonet al., 1997;Fendt and Elstner, 1999;Keppens and Goed-bloed, 2000).

The lesson from these varied simulations seems to bethat all roads lead to a disk field. Thus, the twisting of aclosed magnetic field that initially threads the disk beyondthe co-rotation radius rapidly inflates the field and then dis-connects it from the star, thereby producing a disk-field line.Likewise, simulations including dynamo generated fields inthe disk lead to a state that resembles our initial setup (see§4). Thus, from the numerical point of view, a “fixed-disk”simulation is general and useful.

The setup requires five physical quantities to be speci-fied at all points of the disk surface at all times (see OPIfor all details). These are the disk densityρ(r0); compo-nents of the vertical and toroidal magnetic field,Bz(r0)andBφ(r0); and velocity components in the disk,vz(r0)andvφ(r0), wherer0 is the radius (cylindrical coordinatesare adopted with the disk located atz = 0; r0 = r(z = 0)).The remaining field componentBr(r0) is determined by thesolenoidal condition, while the radial inflow speed throughthe disk is neglected (vr(r0) ' 0) since it is far smaller thanthe sound speed in a real disk. The model is described byfive parameters defined at the inner disk radiusri, three ofwhich describe the initial corona. The two additional pa-rameters describe the disk physics and areνi, which scalesthe toroidal field in the diskBφ = νi × (ri/r0), and the

5

0 20 40 6005

1015

0 20 40 6005

1015

0 20 40 6005

1015

Initial

0 20 40 6005

1015

Final

0 20 40 6005

1015

0 20 40 6005

1015

0 20 40 6005

1015

0 20 40 6005

1015

0 20 40 6005

1015

0 20 40 6005

1015

0 20 40 6005

1015

0 20 40 6005

1015

0 20 40 6005

1015

0 20 40 6005

1015

0 20 40 6005

1015

0 20 40 6005

1015

Fig. 1.— Left panels: initial magnetic field configurations forwinds withµ = 0 (OPI),µ = −0.25 (BP),µ = −0.5 (PP), andµ = −0.75 (steep) Right panels: final magnetic field configura-tions (att = 400) for each case, with Alfvén points (filled circles)and fast magneto-sonic points (stars) marked. Note the moreopenmagnetic – and streamline – structures asµ goes down. (Adaptedfrom PRO).

(subsonic) injection speed of material from the disk into thebase of the corona,vinj = vz(r0)/vφ(r0) ' 0.001. Sim-ulations were typically run with(500 × 200) spatial zonesto resolve a physical region of(80ri × 20ri) in the z andr directions, respectively. A resolution of 10 zones perriprovides enough dynamical range to accurately follow thesmooth acceleration above the disk surface. The simula-tions were run up to400ti (whereti is the Kepler time foran orbit at the inner edge of the disk).

A series of magnetic configurations is shown in Fig. 1where the disk field is modeled asBz(r0, 0) ∝ r

µ−10 . The

initial configurations range from the rather well collimated(such as the potential configuration of OP97,µ = 0.0; andthe Blandford- Payne configuration,µ = −0.25), to the ini-tially more open configurations of Pelletier-Pudritz (PP92;µ = −0.5), and steeper (µ = −0.75).

A good example of a configuration that evolves into astationary jet was studied by OPI and is shown in the up-per panel of Fig. 1). The simulation shows the existenceof an acceleration region very close to the disk surface.The acceleration from the disk occurs by a centrifugal ef-fect whereby, at some point along sufficiently inclined fieldlines, centrifugal force dominates gravity and gas is flungaway like beads on a wire. Thus, a toroidal field componentis created because the field lines corotate with the under-lying disk. The inertia of the matter in the flow region ulti-mately forces the field to fall behind the rotation of the disk,

which produces the toroidal field component. More pre-cisely, beyond the Alfvén surface (shown as dots in the righthand panels in Fig. 1), the hoop stress induced by the self-generatedBφ eventually dominates, which provides the col-limation. The ratio of the toroidal to poloidal magnetic fieldalong illustrative field lines (e.g. Figs. 3 in OPS) clearlyshows that the predominant magnetic field in jets beyondtheir fast magnetosonic (FM) surfaces is the toroidal fieldcomponent. The gas is eventually collimated into cylindersparallel to the disk’s axis starting at the Alfvén surface todistances much beyond the FM surface (see Fig. 1).

The final velocities achieved by such winds are of theorder of 2 times the Kepler velocity, along a given field line(e.g., Fig. 3 in OPS), which translates to roughly 100-300km s−1 for a standard YSO. The general trend is that thejet solutions, dominated mainly by the poloidal kinetic en-ergy, are very efficient in magnetically extracting angularmomentum and energy from the disk, as confirmed by sim-ulations performed by other groups. In 1000 years, for ex-ample, and for a standard YSO, the simulations imply thatthe disk winds can carry a total energy of3 × 1043 ergs,sufficient to produce the observed molecular outflows

In OPII an initial vertical field configuration, was used.The simulations show that the strong toroidal magnetic fieldgenerated recollimates the flow towards the disk’s axis and,through MHD shocks, produces knots. The knot generatoris located at a distance of aboutz ' 8ri from the surfaceof the disk (OPS). Knots propagate down the length of thejet at speeds less than the diffuse component of the outflow.The knots are episodic, and are intrinsic to the jet and notthe accretion disk, in this calculation. It was later discov-ered that the steady state solutions could become episodicby varying the mass-load at the base of the jet (see OPIII).

A different initial state was used byKrasnopolsky, Li,and Blandford(1999), who introduce and maintain through-out the simulation, a strong outflow on the outflow axis.Otherwise, they choose an initial disk field distribution thatis the same as PP92, a mass flux densityρvz ∝ r

3/2o , so that

k = const.. The values ofBr andBφ at the disk surfacewere not fixed in their simulations. (This does not guar-antee that the disk or the boundary at the base will remainKeplerian over time. The fixedBφ ∝ r

−10 profile adopted

in the OP setup ensures exactly that - see the discussionaround Eq (3.46) in OPI). To evade the problems with theinitial setup mentioned above, these authors continuouslylaunch a cylindrical wind on the axis. This imposed jet in-troduces currents which could significantly affect the sta-bility and collimation of the disk wind. Their disk massloading would predict that the disk wind should not be wellcollimated (see§2.4), whereas their simulation does appearto collimate. This suggests that their on-axis jet may beplaying a significant role in the simulation results.

3.2. The role of mass loading

The mass-loadk, defined in equation (7), can be estab-lished by varying; the coronal density profile while keep-

6

ing ρvp constant (Anderson et al., 2005), the disk rota-tional profile sincevp = vinjvφ (e.g.,Vitorino et al., 2002),the distribution of the poloidal magnetic field on the disk(PRO), or the distribution of both the disk magnetic fieldand the mass flow profile (Fendt2005).

The prediction that the mass load determines the colli-mation of the jet was tested in simulations by PRO. Fig. 1shows that simulations withµ = 0, −0.25 collimate intocylinders, while theµ = −0.5 case (PP92) transitions to-wards the wide angle flow seen in theµ = −0.75 simu-lation. These results confirm the theory laid out in§2.4.We also note that each of the simulations mentioned aboveresults in a unique rotation profile of the jet that might inprinciple be observable. Going from the potential case tothe Pelletier-Pudritz configuration, the radial profiles oftherotational velocity of the jets scale as power laws

vφ(r,∞) ∝ ra (18)

wherea = −0.76; −0.66; −0.46 respectively. The ob-servation of a rotational profile would help pick out a uniquemass loading in this model (PRO).

The mass loading also affects the time-dependent be-haviour of jets. As the mass load is varied a transitionfrom stationary, to periodic and sometimes to a discontin-uous dying jet occurs. For example, in some of the sim-ulations it was found that low mass loads for jets lead torapid, episodic behaviour while more heavily mass-loadedsystems tend to achieve stationary outflow configurations(see OPIII).

It is interesting that the simulations ofAnderson et al.(2005) did not find any non-steady behavior forlow massloading but instead find an instability forhighmass loading.They suggest that the origin of the non-steady behavior forhigh mass loading is that the initially dominant toroidal fieldthey impose could be subject to the kink instability (e.g.,Cao and Spruit, 1994). The very large mass-loads, andlarge injection speeds (vinj = (0.01, 0.1)) they use drive aninstability probably related to excessive magnetic braking.Some of the main differences with OP are: (i) the use of thenon-equilibrium set up ofKrasnopolsky et al.noted above,which introduces a strong current along the axis that couldstrongly affect the stability properties of these outflows;and(ii) the large injection speeds make the sonic surface tooclose to the disk (the condition thatcs � vK is not satis-fied) and does not provide enough dynamical range for thegas launched from the disk to evolve smoothly. Numericalinstabilities reminiscent of what was found byAnderson etal. (2005) were observed by OP and these often disappearwith high enough resolution (i.e. dynamical range). In fact,we suspect that these type of instabilities will disappear al-together when a proper disk is included in the simulations(e.g.Casse and Keppens, 2002).

3.3. 3-Dimensional simulations

The stability of jets is one of the principal remainingchallenges in the theory. It is well known that the purely

toroidal field configurations that are used to help confinestatic, 3-D, tokamak plasmas are unstable (e.g.,Roberts,1967; Bateman, 1980). The resulting kink, or helical(m = 1) mode instability derived from a 3-D linear sta-bility analysis is powered by the free energy in the toroidalfield, namelyB2φ/8π (Eichler, 1993). Why are real 3-D jetsso stable over great distances in spite of the fact that theyare probably threaded by strong toroidal fields?

The 3D simulations needed to investigate the importanceof kink modes are still rare. Early attempts includeLucekand Bell(1996) andHardee and Rosen(1999; see also ref-erences therein) who performed 3-D simulations of “equi-librium” jets, and found that these uniform, magnetizedjet models remain Kelvin-Helmholtz (K-H) stable to low-order, surface helical and elliptical modes (m = 1, 2), pro-vided that jets are on average sub-Alfvénic. This is in ac-cord with the prediction of linear stability analysis. How-ever, most configurations for jet simulations use ratheradhocprescriptions for the initial toroidal field configurationso that it is difficult to assess how pertinent the results areto the case of a jet that establishes its own toroidal field asthe jet is accelerated from the accretion disk. In general, theavailable analytic and numerical results for the stabilityofsimple jets show that the fastest growing modes are of K-H type. These K-H instabilities are increasingly stabilizedfor super-Alfvénic jets, asMFM is increased much beyondunity. It is also generally known that sub-Alvénic jets arestable. Taken together, these results suggest that 3-D jetsare the most prone to K-H instabilities a bit beyond theirAlfvén surface, a region wherein their destabilizing super-Alfvénic character cannot yet be offset by the stabilizingeffects engendered at large super FM numbers.

There have, as yet, only been a few attempts to simu-late 3D disk winds. Contrary to what may be intuitive, itis inadvisable to perform these 3-D simulations in cylindri-cal coordinates. For one thing, special treatment must begiven to the “wedge zones” that abut thez-axis (no longera symmetry axis in 3D), and velocities that pass throughthez-axis pose a very difficult numerical problem. Second,even with such technical details in hand, plane waves arebadly disrupted on passing through thez-axis, and this pro-vides an undesirable bias to what should be an unbiased (noaxis should be preferred over another) 3D calculation. For-tunately, using Cartesian coordinates to simulate cylindersis feasible with careful setups as has been demonstrated inOuyed et al.(2003; OCP); andOuyed(2003). These au-thors used a 3D version of the OPI setup.

The central result of this study is that jets survive thethreatening non-axisymmetric kink (m = 1) mode. Thesimulated jets maintain their long-term stability throughaself-limiting process wherein the average Alfvénic Machnumber within the jet is maintained to order unity. Thisis accomplished in at least two ways: (i) poloidal magneticfield is concentrated along the central axis of the jet form-ing a “backbone” in which the Alfvén speed is sufficientlyhigh to reduce the average jet Alfvénic Mach number tounity, and (ii) the onset of higher order Kelvin-Helmholtz

7

Fig. 2.—Cylindrical Jets – The upper panels show snapshots of20 magnetic field lines of the 3D simulations performed in OCP,** at t = 50, 130, 210, and240. The two central magnetic fieldlines (dotted lines) originate on the central compact object (illus-trated by the semi-sphere to the left). The disk axis is alongthediagonal of the frame (on a45◦ angle). Notice the violent jet be-havior in the in-between frames when the kink mode appears. Thejet eventually aligns itself with the original disk’s axis acquiring acylindrically collimated shape. Corkscrew Jets – The lowerpanelis a map of the jet column density att = 320 for a simulationwith a vφ profile, imposed at the accretion disk, that is differentfrom the simulation shown in the upper four panels (see OCP).The jet has settled into a quasi-steady state structure in the shapeof a “corkscrew”; colors are arranged spectrally from from blueto red to represent low and high values of the density. The disk(not visible in this rendering because boundary values are not in-cluded in the line-of-sight integrations) is on the left-hand side ofthe image, and outflow is from left to right. (Adapted from OCP).

“flute” modes (m ≥ 2) reduce the efficiency with whichthe jet material is accelerated, and transfer kinetic energy ofthe outflow into the stretched, poloidal field lines of the dis-torted jet. This too has the effect of increasing the Alfvénspeed, and thus reducing the Alfvénic Mach number. Thejet is able to survive the onset of the more destructivem = 1mode in this way. It was further discovered that jets go intoalternating periods of low and high activity as the disappear-

ance of unstable modes in the sub-Alfvénic regime enablesanother cycle of acceleration to super-Alfvénic speeds. Theperiod of the episodic (low vs. high) behavior uncoveredin the case of the wobbling jet (upper four panels in Fig. 2)is of the order of200 ti. For a typical young stellar objectof mass0.5M� and forri ∼ 0.05 AU (i.e. ti ∼ 0.9 day),this would correspond to a minimum period of roughly180days. This is consistent with the temporal variations of theobserved YSOs outflow velocities which appears to occuron timescales of a few years (Woitas et al.2002).

In the process of taming the effects of the kink mode,jets settle into relatively stable end-states involving eithercorkscrew (lower panel in Fig. 2), or wobbling types ofstructure (upper four panels in Fig. 2). The difference inthe two type of jets can be traced to the difference in thevφ profiles imposed at the accretion disk (see OCP for moredetails). This trend has been uncovered in 3D simulationsof large scale jets where wiggled jets have also been ob-served after the jets survived the destructive era (Uchida etal., 1992;Todo et al., 1993;Nakamura et al., 2001;Naka-mura and Meier, 2004). For completeness, we should men-tion the recent similar simulations performed byKigure andShibata(2005) which included the disk self-consistently.These studies also show that non-axisymmetric perturba-tions dominate the dynamics. However, these simulationswere only run for very short times, never exceeding 2 inner-disk orbital periods due to numerical obstacles reported bythe authors. In OCP simulations, it was shown that theCartesian grid (also used byKigure and Shibata) inducesartificial modes in the early stages of the simulations andonly after many orbital periods that these become washedout before the real modes enter the dynamics.

It has been suggested that 3D jets might avoid a kink in-stability by settling into a minimum-energy “Taylor state”where the jet maintains comparable poloidal and toroidalfield components (Königl and Choudhuri, 1985). This al-ternative is does not pertain to our simplified simulationssince we do not allow for internal magnetic energy dissipa-tion.

We sum up this section by noting that a large body ofdifferent simulations have converged to show that jets; (i)are centrifugally driven in broad agreement with theoreticalwork on centrifugally driven outflows; (ii) are collimated bythe “hoop stress” engendered by their toroidal fields; (iii)achieve two types of configuration depending on the massloading that takes place in the underlying accretion disc –those that achieve collimation towards a cylinder, and thosethat have a wide-angle structure; (iv) achieve two types ofregimes depending on the mass loading that takes place inthe underlying accretion disc – those that achieve a station-ary state, and those that are episodic; (v) achieve stability bya combination of MHD mode coupling and “back-bone” ef-fect leading to a self-regulating mechanism which saturatesthe instabilities; and (vi) achieve different morphologieswhich range from cylindrical, wobbling and cork-screw de-pending on the profile ofvφ imposed on the underlying ac-cretion disk.

8

4. COUPLED DISK-JET EVOLUTION

While it is physically useful to regard the accretion diskas a boundary condition for the jet, this ignores critical is-sues such as the self-consistent radial distribution of themass loading, or of the threading magnetic field across thedisk. These, and other degrees of freedom can be found bysolving the combined disk-outflow problem, to which wenow turn.

Significant progress on the launch of disk winds has beenmade by theoretical studies of a restricted class of self-similar 2D models (e.g.Wardle and K̈onigl, 1993;Li, 1995,1996; Ferreira, 1997; Casse and Ferreira, 2000). Sincecentrifugally driven outflows can occur for completely coldwinds (BP82), the models have examined both “cold” and“warm” (i.e., with some kind of corona) conditions. In theformer class, material from some height above the disk mid-plane must move upwards and eventually be acceleratedoutwards in a jet. Therefore, there must be a direct linkbetween the physics of magnetized disks, and the origin ofoutflows.

Calculations show that in order to match the propertiesof jets measured byBacciotti et al.(2000), warm wind so-lutions are preferred, wherein a disk corona plays a centralrole. In this situation, gas pressure imbalance will assistwith feeding the jet. A warm disk corona is expected ongeneral grounds because it is a consequence of the MRI, asthe vertically resolved disk simulations ofMiller and Stone(2000) have shown.

A semi-analytic model of the radial and vertical diskstructure, that includes self-consistently the outflow, hasalso been presented byCampbell(2003). These solutionsdemonstrate explicitly that the outflow contributes to theloss of angular momentum in the disk by channeling italong field lines into the outflow. For self-consistent so-lutions to be possible, the turbulent Mach number has to bebetween 0.01 and 0.1.

Models of outflows and jets generally assume ideal (non-resistive) MHD. However, inside the disk non-ideal effectsmust become important, because the accreting matter wouldotherwise never be able to get onto the field lines that threadthe disk and connect it with the outflow. In recent two-dimensional models,Casse and Keppens(2002, 2004) as-sumed a resistivity profile, analogously to the Shakura andSunyaev prescription. This assumes the presence of an un-derlying turbulence within the disk. Only fairly large val-ues of the correspondingαSS parameter of around 0.1 havebeen used. The subscript SS refers toShakura and Sunyaev(1973), who were the first to introduce this parameter. Inall cases the system evolves to a steady equilibrium. Us-ing resistive simulations in a different context,Kuwabaraet al. (2005) demonstrated that a substantial amount of en-ergy can be transported by Poynting flux if the poloidal fieldfalls off with distance no faster thanr−2. Otherwise, thefast magnetosonic point is located closer to the Alfvén pointand the jet will be dominated by kinetic energy, which is thecase in the simulations ofCasse and Keppens(2004).

The general stability of disk–outflow solutions is stillbeing debated and the result may depend on the detailedassumptions about the model. In the solutions discussedhere the accretion stress comes entirely from the large scalemagnetic field rather than some small scale turbulence. Asemphasized byFerreira and Casse(2004), real disks havea turbulent viscosity just as they have turbulent magneticdiffusivity or resistivity. However, if the accretion stressdoes come entirely from the large scale magnetic field, thewind–driven accretion flows may be unstable (Lubow et al.,1994). Königl (2004) has shown recently that there are infact two distinct solution branches–a stable and an unstableone. He argues that real disk/wind systems would corre-spond to the stable branch.

Finally, the idea of a gently flared accretion disk is likelyto be merely the result of the modeler’s simplification ratherthan observational reality. Indeed, accretion disks can bewarped due to various instabilities which can be drivenby radiation from the central object (e.g.Pringle, 1996),or, more likely, by the outflow itself (Schandl and Meyer,1994). If a system is observed nearly edge-on, a warp in theaccretion disk produces periodic modulations of the lightcurve. AsPinte and Ḿenard (2004) have demonstrated,this may be the case in AA Tau. Observations frequently re-veal major asymmetries in bipolar outflows, which may betraced back to the corresponding asymmetries in the diskitself. The possible causes for these asymmetries may beeither an externally imposed asymmetry such as one-sidedheating by a nearby OB association, or an internal symme-try breaking of the disk-wind solution as a result increasedrotation. Examples of the latter are familiar from the studyof mean field dynamo solutions of accretion disks (Torkels-son and Brandenburg, 1994).

4.1. Dead zones

So far it has been assumed that the magnetic field is wellcoupled to the disk. Most of the envelope of an accretiondisk is ionized by a combination of cosmic rays, as wellas X-rays from the central YSO so that good coupling be-tween the charged particles and the magnetic field can oc-cur. However, the deeper layers of the disk are stronglyshielded from these ionizing agents and the degree of ion-ization plumets.

This dense layer, which encompasses the bulk of thedisk’s column density, cannot maintain a sufficiently highelectron fraction, and is referred to as the dead zone (Gam-mie, 1996). It is the poorly ionized region within whichthe magneto-rotational instability (MRI) fails to grow as aconsequence of the diffusivity of the field. Recent workof Fleming and Stone(2003) shows that, although the lo-cal Maxwell stress drops to negligible values in the deadzones, the Reynolds stress remains approximately indepen-dent of height and never drops below approximately 10% ofthe maximum Maxwell stress, provided the column densityin that zone is less that 10 times the column of the activelayers. The non-dimensional ratio of stress to gas pressure

9

is just theShakura and Sunyaev(1973) viscosity parame-ter, αSS. Fleming and Stone(2003) find typical values ofa few times10−4 in the dead zones and a few times10−3

in the MRI-active layers.Inutsuka and Sano(2005) havequestioned the very existence of dead zones, and proposedthat the turbulent dissipation in the disk provides sufficientenergy for the ionization. On the other hand, their calcu-lation assumes a magnetic Reynolds number that is muchsmaller than expected from the simulations (seeMatsumuraand Pudritz, 2006; MP06).

The radial extent of the dead zone depends primarilyupon the disk column density as well as effects of grains andradiation chemistry (e.g.,Sano et al., 2000;Matsumura andPudritz, 2005, MP05; and MP06). Inside the dead zone themagnetic Reynolds number tends to be below a certain crit-ical value that is somewhere between 1 and 100 (Sano andStone, 2002), making MRI-driven turbulence impossible.Estimates for the radial extent of the dead zone range from0.7–100 AU (Fromang et al., 2002), to 2–20 AU in calcu-lations bySemenov et al.(2004). ForChiang et al.(2001)models of disks that are well constrained by the observa-tions, and whose surface density declines asΣ ∝ r−3/20 ,MP05 find that the extent of the dead zone is robust – ex-tending out to 15 AU typically, and is fairly independent ofthe ionizing environment of the disk.

For smaller radii, thermal and UV ionization are mainlyresponsible for sustaining some degree of ionization. Thesignificance of the reduced value ofαSS in the dead zonesis that it provides a mechanism for stopping the inward mi-gration of Jupiter-sized planets (MP05, MP06). Jets canstill be launched from the well-coupled surface layer abovethe dead zone (e.g.Li, 1996;Campbell, 2000).

When the MRI is inactive in the body of the disk, alter-native mechanisms of angular momentum transport are stillpossible. In protostellar disks there are probably at leasttwo other mechanisms that might contribute to the accre-tion torque: density waves (Różyczka and Spruit, 1993), andthe interaction with other planets in the disk (Goodman andRafikov, 2001).

A more controversial alternative is to drive turbulence bya nonlinear finite amplitude instability (Chagelishvili et al.,2003). WhileHawley et al.(1999) have given general ar-guments against this possibility,Afshordi et al.(2005) andother groups have continued investigating the so-called by-pass to turbulence. The basic idea is that successive strongtransients can maintain a turbulent state in a continuouslyexcited manner.Lesur and Longaretti(2005) have recentlybeen able to quantify more precisely the critical Reynoldsnumber required for instability. They have also highlightedthe importance of pressure fluctuations that demonstratethat the general argument byBalbus et al.(1996) is insuffi-cient.

Finally, the role of vertical (convectively stable) densitystratification in disks and the possibility of the so-calledstrato-rotational instability has been proposed as a possi-ble mechanism for disk turbulence (Dubrulle et al., 2005).

This instability was recently discovered byMolemaker etal. (2001) and the linear stability regime was analyzed byShalybkov and R̈udiger (2005). However, the presence ofno-slip radial boundary conditions that are relevant to ex-periments and used in simulations are vital. Indeed, the in-stability vanishes for an unbounded regime, making it irrel-evant for accretion disks (Umurhan2006).

4.2. Advected vs. dynamo generated magnetic fields

One of the central unresolved issues in disk wind theoryis the origin of the threading magnetic field. Is it dragged inby the gravitational collapse of an original magnetized core,or is it generatedin situby a dynamo of some sort?

The recent direct detection of a rather strong of a truedisk field of strength 1 kG at 0.05 AU in FU Ori, pro-vides new and strong support for the disk wind mecha-nism (Donati et al. 2005). The observation techniqueuses high-efficiency high-resolution spectropolarimetryandholds much promise for further measurements. This workprovides excellent evidence that distinct fields exist in disksin spite of processes such as ambipolar diffusion whichmight be expected to reduce them.

Turbulence effects. Accretion disks are turbulent–certainly in the well-coupled lower corona–and they areintrinsically three-dimensional and unsteady. Questionsre-garding the stability of disks concern therefore only themean (azimuthally averaged) state, where all turbulent ed-dies are averaged out. The averaged equations used to ob-tain such solutions incorporate a turbulent viscosity. Inthis connection we must explain that the transition from amacroscopic viscosity to a turbulent one is more than justa change in coefficients, because one is really entering anew level of description that is uncertain in many respects.For example, in the related problem of magnetic diffusionit has been suggested that in MRI-driven turbulence, thefunctional form of of turbulent magnetic diffusion may bedifferent, such that it operates mainly on the system scaleand less efficiently on smaller scales (Brandenburg andSokoloff, 2002). It is therefore important to use direct sim-ulations to investigate the stability in systems that mightbeunstable according to a mean field description. There is nogood example relevant to protostellar disks, but related ex-perience with radiatively dominated accretion disks, whereit was found that the so-called Lightman–Eardly instabilitymight not lead to the breakup of the disk (Turner et al.,2004), should provide enough reason to treat the mean-fieldstability problem of protostellar disks with caution. Also,one cannot exclude that circulation patterns found in sim-ulations of the two-dimensional mean-field equations (e.g.,Kley and Lin, 1992) may take a different form if the fullyturbulent three-dimensional problem was solved.

The turbulent regions of disks will generally be capableof dynamo action – wherein an ordered field is generatedby feeding on the energy that sustains the turbulence. Thesetwo aspects are in principle interlinked, as has been demon-strated some time ago using local shearing box simulations.

10

These simulations show not only that the magnetic field isgenerated from the turbulence by dynamo action, but alsothat the turbulence itself is a consequence of the magneticfield via the MRI (Brandenburg et al., 1995;Stone et al.,1996).

In comparing numerical dynamos with observational re-ality it is important to distinguish between two differenttypes of dynamos: small scale and large scale dynamos.The names refer primarily to the typical scale of the field.Both types of dynamos have in general a turbulent com-ponent, but large scale dynamos have an additional com-ponent on a scale larger than the typical scale of the tur-bulence. Physically, this can be caused by the effects ofanisotropies, helicity, and/or shear. These large scale dy-namos are amenable to mean field modeling (see below).Small scale dynamos, on the other hand, tend to be quiteprominent in simulations, but they are not described bymean field models. However, small scale dynamos could bean artefact of using unrealistically large magnetic Prandtlnumbers in the simulations. Protostellar disks have mag-netic Prandtl numbers,ν/η, around10−8, so the viscos-ity of the gas,ν, is much smaller than the diffusivity ofthe field, η; see Table 1 ofBrandenburg and Subrama-nian (2005). High magnetic Prandtl numbers imply that thefield is advected with the flow without slipping too much,whereas for low numbers the field readily slips significantlywith respect to the flow. At the resistive scale, where thesmall scale dynamo would operate fastest, the velocity isstill in its inertial range where the spatial variation of thevelocity is much rougher than for unit magnetic Prandtlnumber (Boldyrev and Cattaneo, 2004). This tends to in-hibit small scale dynamo action (Schekochihin et al., 2005).In many simulations, especially when a subgrid scale pre-scription is used, the magnetic Prandtl number is effectivelyclose to unity. As a consequence, the production of smallscale field may be exaggerated. It is therefore possible thatin real disks large scale dynamo action is much more promi-nent than what is currently seen in simulations.

In mean field models, only the large scale field is mod-eled. In addition to a turbulent magnetic diffusivity thatemerges analogously to the turbulent viscosity mentionedabove, there are also non-diffusive contributions such as theso-calledα effect wherein the mean electromotive force canacquire a component parallel to the mean field,αdynB; seeBrandenburg and Subramanian(2005) for a recent review.For symmetry reasons,αdyn has on average opposite signsabove and below the midplane. Simulations ofZiegler andRüdiger(2000) confirm the unconventional negative sign ofαdyn in the upper disk plane, which was originally found inBrandenburg et al.(1995). This has implications for theexpected field parity which is expected to be dipolar (anti-symmetric about the midplane); seeCampbell et al.(1998)andBardou et al.(2001). Although a tendency toward dipo-lar fields is now seen in some global accretion disk simula-tions (De Villiers et al., 2005), this issue cannot as yet beregarded as conclusive.

Outflows from dynamo-active disks. Given that the

disk is capable of dynamo action, what are the relativeroles played by ambient and dynamo-generated fields? Thisquestion has so far only been studied in limiting cases. Thefirst global simulations were axisymmetric, in which casedynamo action is impossible by theCowling (1934) the-orem. Nevertheless, such simulations have demonstratedquite convincingly that a jet can be launched by windingup the ambient field and driving thereby a torsional Alfvénwave (Matsumoto et al., 1996). These simulations are scaleinvariant and, although they were originally discussed inthe context of active galaxies, after appropriate rescaling,the same physics applies also to stellar disk outflows.

When the first three-dimensional simulations becameavailable an immediate issue was the demonstration thatMRI and dynamo action are still possible (Hawley, 2000).Although the simulations were applied to model blackwhole accretion disks, many aspects of black hole accre-tion are sufficiently generic and carry over to protostel-lar disks as well. These simulations showed that the dy-namo works efficiently and produces normalized accretionstresses (αSS ≈ 0.1) that exceed those from local simula-tions (αSS ≈ 0.01). The global simulations did not at firstseem to show any signs of outflows, but this was mainlya matter of looking at suitable diagnostics, which empha-sizes the tenuous outflows rather than the much denser diskdynamics (De Villiers et al., 2005). Although these simula-tions have no ambient field, a large scale field develops inthe outflow region away from the midplane. However, it isdifficult to run such global simulations for long enough toexclude a dependence of the large scale field on the initialconditions. The outflows are found to be uncollimated suchthat most of the mass flux occurs in a conical shell withhalf-opening angle of25–30◦ (De Villiers et al., 2005).

These very same properties are also shared by mean fieldsimulations, where the three-dimensional turbulent dynam-ics is modeled using axisymmetric models (von Rekowskiet al., 2003). In these simulations (Fig. 3), where the fieldis entirely dynamo-generated, there is an uncollimated out-flow with most of the mass flux occurring in a conicalshell with a half-opening angle of about25–30◦, just likein the three-dimensional black hole simulations. The fieldstrength generated by dynamoc action in the disk is foundto scale asBp ∝ r−2. This is very similar to the scaling forpoloidal field strength found in collapse simulations (seeBanerjee and Pudritz2006).

The conical outflows discussed above have tentativelybeen compared with the observed conical outflows inferredfor the BN/KL region in the Orion nebula out to a dis-tance of 25–60 AU from its origin (Greenhill et al., 1998).One may wonder whether collimated outflows are onlypossible when there is an ambient field (discussed in§3).This conclusion might be consistent with observations ofthe Taurus-Auriga molecular cloud.Ménard and Ducĥene(2004) found that, although T Tauri stars are oriented ran-domly with respect to the ambient field, there are no brightand extended outflows when the axis is perpendicular to theambient field. Note thatSpruit and Uzdensky(2005) have

11

Fig. 3.— Outflow from a dynamo active accretion diskdriven by a combination of pressure driving and magneto-centrifugal acceleration. The extent of the domain is[0, 8] × [−2, 30] in nondimensional units (correspondingto about[0, 0.8] AU × [−0.2, 3] AU in dimensional units).Left panel: velocity vectors, poloidal magnetic field linesand gray scale representation of the specific enthalpyh inthe inner part of the domain. Right panel: velocity vec-tors, poloidal magnetic field lines and normalized specificenthalpyh/|Φ| in the full domain. [Adapted fromvonRekowski et al.(2003)].

argued that the efficiency of dragging an ambient field to-ward the star have been underestimated in the past, providedthe field segregates into many isolated flux bundles.

In summing up this section we note that all magneticdisk models produce outflows of some form. However, thedegree of collimation may depend on the possibility of anambient field. An important ingredient that seems to havebeen ignored in all combined disk-jet models is the effectof radiation. There are also many similarities between theoutflows in models with and without explicitly including thedisk structure. Although these simulations are non-ideal (fi-nite viscosity and resistivity), the various Lagrangian invari-ants (mass loading parameter, angular velocity of magneticfield lines, as well as the angular momentum and Bernoulliconstants) are still very nearly constant along field lines out-side the disk.

5. JETS AND STAR-DISK INTERACTION

In this section we review the recent development con-cerning the interaction of young stellar magnetosphereswith the surrounding accretion disk and the contributionthat these processes may have to the formation of jets.

Perhaps the most important clue that star-disk interac-

tion is an important physical process comes from the obser-vation that many YSOs are observed to spin at only a smallfraction (' 10%) of their break-up speed (e.g.,Herbst etal., 2002). They must therefore undergo a significant spin-down torque that almost cancels the strong spin-up that aris-ing from gas accretion from the inner edge of the Kepleriandisk.

Three possible spin-down torques have been proposed:(i) disk-locking – a magnetospheric connection between thedisk and the star which dumps accreted angular momen-tum back out into the disk beyond the co-rotation radius(Königl 1991); (ii) an X-wind – a centrifugally driven out-flow launched from the very inner edge of the accretiondisk which intercepts the angular momentum destined forthe star (see chapter byShang et al., this volume); and (iii)an accretion-powered stellar wind – in which the slowly ro-tating, central star drives a massive (Ṁw,∗ ' 0.1Ṁa), stel-lar wind that carries off the bulk of the angular momentumas well as a fraction of the energy, that is deposited intothe photosphere by magnetospheric accretion (Matt and Pu-dritz, 2005b). The last possibility implies that the centralobject in an accretion disk also drive an outflow, which isin addition to the disk wind. Separating the disk wind fromthe outflow from the central object would be difficult, butthe shear layer that separates them might generate instabil-ities and shocks that might be diagnostics.

The strength of the star-disk coupling depends on thestrength of the stellar magnetic field. Precise stellar mag-netic field measurements exist for nine TTSs, while forseven other stars statistically significant fields have beenfound (seeSymington et al., 2005b and references therein).Zeeman circular-polarization measurements in He I of sev-eral classical TTs give direct evidence of kG magnetic fieldswith strong indication of considerable field variation on thetime scales of years (Symington et al.2005b). For a sam-ple of stars a magnetic field strength up to 4 kG fields couldbe derived (DF Tau, BP Tau), as for others, notably the jetsource DG Tau, no significant field could be detected at thetime of observation.

5.1. Basic processes

Variable accretion from a tipped dipole. A centraldipolar field inclined to the rotation axis of star and diskmay strongly disturbs the axisymmetry of the system. Pho-tometric and spectroscopic variability studies of AA Taugive evidence for time-dependent magnetospheric accretionon time scales of the order a month. Monte-Carlo modelingshows that the observed photo-polarimetric variability mayarise by warping of the disk that is induced by a tipped mag-netic dipole has been reproduced (O’Sullivan et al., 2005).More general investigations of the warping process by nu-merical simulations show that the warp could evolve intoa steady state precessing rigidly (Pfeiffer and Lai2004).Disks can be warped by the magnetic torque that arisesfrom the a slight misalignment between the disk and star’srotational axis (Lai 1999). This disk warping mechanism

12

Fig. 4.—Slice through a funnel stream. Density contours fromρ = 0.2 to ρ = 2.0. The density of the disk corona isρ =0.01 − 0.02 Selected magnetic field lines. The rotational axis andthe magnetic moment are indicated asΩ andµ. [Adapted fromRomanova et al.(2004).]

may also operate in the absence of a stellar magnetosphereas purely induced by the interaction between a large-scalemagnetic field and the disk electric current and, thus, maylead to the precession of magnetic jets/outflows (Lai 2003).

Three-dimensional radiative transfer models of the mag-netospheric emission line profile based on the warped diskdensity and velocity distribution obtained by numericalMHD simulations give gross agreement with observationswith a variability somewhat larger than observed (Syming-ton et al., 2005a).

Magnetic flux. Compared to the situation of a puredisk magnetic field, the magnetic field of the star may addsubstantial magnetic flux to the system. For a polar fieldstrengthB0 and a stellar radiusR?, the large scale stellardipolar field (ignoring angular variations)

Bp,?(r) ' 40 G

(

B01 kG

)(

r

3 R?

)−3

(19)

has to be compared to the accretion disk poloidal magneticfield which is provided either by a disk dynamo or by advec-tion of ambient interstellar field, both limited by equiparti-tion arguments,

Bp,disk < Beq(r) = 20 Gα−1/2

(

Ṁa10−6 M� yr−1

)1/2

·

(

M?M�

)1/4 (H/r

0.1

)−1/2(r

10 R�

)−5/4

.(20)

This flux will not remain closed, but will inflate and open upas magnetic field lines are sheared and extract gravitationalpotential energy from the accreting flow (e.g.,Uzdensky etal., 2002;Matt and Pudritz, 2005a). These field lines there-fore effectively become a disk field, and therefore followthe processes of disk wind production we have already dis-cussed.

The additional Poynting flux that threads the disk mayassist the jet launching by MHD forces and may serve as

an additional energy reservoir for the kinetic energy of thejet implying greater asymptotic jet speed (Michel scaling;Michel1969,Fendt and Camenzind1996).

Disk locking vs. stellar winds.The spin of the star willdepend on how angular momentum arriving from the disk,is dealt with. In the magnetic “disk locking” picture, thethreading field of the disk will re-arrange the global angularmomentum budget. The torque on the star by the accretionof disk matter is

τa = Ṁa (GM?rin)1/2 (21)

(e.g.,Matt and Pudritz, 2005a), with the disk accretion rateṀacc, the stellar massM? and the disk inner radiusrin in-side the corotation radiusrco = (GM?)1/3Ω

−2/3? . On the

other hand if “disk locking” is present, the stellar rotationmay be decelerated by the magnetic torque due to stellarfield lines connecting the star with the accretion disk out-sideRco. The differential magnetic torque acting on a diskannulus ofdr width is

dτmag = r2BφBzdr. (22)

WhileBz in principle follows from assuming a central dipo-lar field, the induction of toroidal magnetic fields is modeldependent (disk resistivity, poloidal field structure). This iswhy recent numerical simulations of dipole-disk interactionthat simultaneously evaluate the poloidal and toroidal fieldcomponents have become extremely valuable.

If indeed the star loses angular momentum to the disk(this is not yet decided by the simulations, see below), bothdisk accretion and jet formation are affected. In order tocontinue accretion, excess angular momentum has to be re-moved from the accreting matter. A disk jet can be an ef-ficient way to do this, as has been supposed in the X-windpicture.

The central stellar magnetic field may launch a strongstellar wind to rid itself of the accreted angular momen-tum. Such an outflow will interact with the surrounding diskwind. If true, observed YSO jets may consist of two com-ponents – the stellar wind and the disk wind, with strengthdepending on intrinsic (yet unknown) parameters. The stel-lar wind (open field lines of stellar magnetosphere) exerts aspin-down torque upon the star of magnitude

Ṁwind,?Ω?r2A = 3 × 10

36 g cm2

s yr

(

Ω?10−5 s−1

)

(23)

·

(

rA30R�

)2(

Ṁwind,?10−9M�yr−1

)

As a historical remark we note that the model topologyof dipole-plus-disk field were introduced for protostellarjetformation more then 20 years ago in the MHD simulationsof Uchida and Shibata(1984). Further investigations con-sidering the detailed physical processes involved in disktruncation and channeling the matter along the dipolar fieldlines byCamenzind(1990),Königl (1991) andShu et al.(1994) resulted in a breakthrough of these ideas to the pro-tostellar jet community.

13

5.2. Numerical simulations of star-disk interaction

The numerical simulation of the magnetospheric star-disk interaction is technically most demanding as the na-ture of the object requires to treat a complex model geom-etry in combination with strong gradients in magnetic fieldstrength, density and resistivity. In general, this may im-ply a large variation in physical times scales for the threecomponents of disk, jet, and magnetosphere, which all haveto be resolved numerically. Essentially, numerical model-ing of the star-disk interaction would require adiffusive andviscous MHD code including radiative transfer. In addi-tion, realistic models for reconnection processes and diskopacity are needed. Then, such a code has to run withhighspatial and temporal resolutionand, at the same time, on aglobal scale enclosing all three components.

Compared to the situation about a decade ago when therewas still onlyUchida and Shibata’s (1984) initial (thoughten years old) simulation available, today huge progress hasbeen made with several groups (and also codes) competingin the field. Early simulations were able to follow the evolu-tion only for a few rotations of the inner disk (note that 100rotations at the co-rotation radius correspond to 0.3 Keple-rian rotations at 10 co-rotation radii) (Hayashi et al., 1996,Miller and Stone, 1997,Goodson et al., 1997). Among theproblems involved is the initial condition of the simulation,in particular the nature of the disk model which could benumerically treated. A steady initial corona will stronglyshear with the Keplerian disk leading to current sheet (thuspressure gradients) along the disk surface. If the simula-tions run long enough, this could be an intermittent feature.However, the danger exists that the artificial current sheetwill fatally destroy the result of the simulation. Applyinga Shakura-Sunyaevα-disk for the initial disk structure, thecode should also considerα-viscosity. Otherwise the initialdisk evolution is not self-consistent (see, e.g.,Goodson etal., 1997).

The next step is to increase the grid resolution and toredo the axisymmetric simulations.Goodson et al.(1999)and Goodson and Winglee(1999) were able to treat sev-eral hundreds of (inner) disk rotations on a global grid of50 AU extension. The main result of these simulations is atwo component flow consisting of a fast and narrow axialjet and a slow disk wind, both launched in the inner partof the disk (r < 1 AU). An interesting feature is that thenarrow axial jet is actually a collimation in density and notin velocity. Close to the inner disk radius repetitive recon-nection processes are seen on time scales of a couple ofrotation periods. The dipolar field inflates and an expand-ing current sheet builds up. After field reconnection alongthe current sheet, the process starts all over again. The os-cillatory behaviour leads to the ejection of axial knots. Onaverage the central dipolar magnetosphere remains presentand loaded with plasma. Forbidden emission line intensitymaps of these simulations have been calculated and allowfor direct comparison with the observations. However, thenumerically derived time and length scales of axial knots

were still too different from what is observed. The magne-tospheric origin of jets (stellar dynamo) in favor of a puredisk origin (“primordial field”) has also been stressed byMatt et al.(2002).

Another approach was taken byFendt and Elstner(1999,2000). In order to be able to perform long-term simula-tions of dipolar magnetospheres interacting with the accre-tion disk, the evolution of the disk structure was neglectedand instead taken as a fixed boundary condition for the out-flow (as in OPI). After 2000 rotations a quasi-steady statewas obtained with a two-component outflow from disk andstar. The outflow expands almost radially without signa-ture of collimation on the spatial scale investigated (20×20inner disk radii). One consequence of this very long simu-lation is that the axial narrow jet observed in other simula-tions might be an intermittent feature launched in the earlyphase of the simulation as the initial field is reconfigured toa new equilibrium. The axial outflow in this simulations ismassive but slow, but tends to develop axial instabilities fora lower mass loading. Clearly, as in this approach the diskstructure has not been taken into account, nothing could besaid about the launching process of the outflow out of thedisk.

In a series of ideal MHD simulations Romanova and col-laborators succeeded in working out a detailed and suffi-ciently stable numerical model of magnetospheric disk in-teraction. They were the first to simulate the axisymmetricfunnel flow from the disk inner radius onto the stellar sur-face (Romanova et al., 2002) on a global scale (Rmax =50Rin) and for a sufficiently long period of time (300 rota-tions) in order to reach a steady state in the accretion fun-nel. The funnel flow with free-falling material builds up as agap opens up between disk and star. The authors further in-vestigated the angular momentum exchange due to the disk“locking” by the funnel flow. Slowly rotating stars seem tobreak the inner disk region and spin up, while rapid rotatorswould accelerate the inner disk region to super-Keplerianvelocity and slow down themselves. Only for certain stel-lar rotational periods a “torque-less” accretion could be ob-served. Strong outflows have not been observed for theparameter space investigated, probably due to the matterdominated corona which does not allow for opening-up thedipolar field.

Further progress has been achieved extending these sim-ulations to three dimensions (Romanova et al., 2003, 2004).For the first time it has been possible to investigate the inter-action of an inclined stellar dipolar magnetosphere with thesurrounding disk. For zero inclination axisymmetric fun-nels are found as in the axisymmetric simulations. For non-zero inclination the accretion splits in two main streams fol-lowing the closest path from the disk to the stellar surface asis seen in Fig. 4. Magnetic braking changes the disk struc-ture in several ways – its density structure (gaps, rings), itsvelocity structure (sub-Keplerian region) and its geometry(warping). The slowly rotating star is spun up by accret-ing matter but this acceleration does only weakly dependon inclination. For completeness we should note that for

14

these simulations a clever approach for the numerical gridhas been applied, the “cubed grid” which does not obey thesingular axes as for spherical coordinates.

The star-disk coupling by the stellar magnetosphere wasalso investigated byKüker et al.(2003). These simulationshave been performed in axisymmetry, but an advanced diskmodel has been applied. Taking into accountα-viscosity, acorresponding eddy magnetic diffusivity and radiative en-ergy transport, the code enables the authors to treat a realis-tic accretiondisk in their MHD simulations. As a result thesimulations could be advanced to very long physical timesteps into a quasi stationary state of the disk evolution. Theauthors show that the commonly assumed 1000 G magne-tosphere is not sufficient to open up a gap in the disk. Un-steady outflows may be launched outside of the corotationradius with mass loss rates of about 10 % of the accretionrate. The authors note that they were unable to detect thenarrow axial jet seen in other publications before (in agree-ment withFendt and Elstner, 2000). Magnetic braking ofthe star may happen in some cases, but is overwhelmedby the accretion torque still spinning up the star (however,these simulations treat the stellar surface as inner boundarycondition).

Using a different approach, the boundary condition ofa stellardipolar magnetosphere or a large scale disk fieldwere relaxed. This allows for the interaction of a stellardipole with adynamo-generated disk magnetic field (vonRekowski and Brandenburg, 2004), or the evolution of adynamo-generated stellar field affected by a surroundingdisk with its own disk dynamo (von Rekowski and Bran-denburg, 2006) to be studied. The results of this workare in agreement with previous studies, and directly provethat a disk surrounding a stellar magnetosphere may actu-ally develop its own magnetic field strong enough to launchMHD outflows. In agreement with results byGoodson andWinglee(1999), accretion tends to be unsteady and alter-nating between connected and disconnected states. Evenfor realistic stellar fields of several hundred gauss the mag-netic spin-down torque is insufficient to overcome the spin-up torque from the accretion flow itself. As shown bythe authors, angular momentum exchange is complex andmay vary in sign along the stellar surface, braking someparts of the star and accelerating others. The dynamo-generated stellar field may reach up to 750 G and switchesin time between dipolar and quadrupolar symmetry, whilethe dynamo-generated 50 G disk field is of dipolar symme-try. In general, the dynamo-generated stellar field is bettersuited to drive a stellar wind. The simulations show that forthese cases stellar wind braking is more efficient than brak-ing by the star-disk coupling, confirming the results ofMattand Pudritz(2005a).

Recent studies of torque-less accretion (Long et al.,2005) compare cases of a (i) weak stellar magnetic fieldwithin a dense corona and (ii) a strong field in a lower den-sity corona. In particular, the angular momentum transportalong the open and closed magnetic field lines is investi-gated pointing out the importance of field line reconnection

on quasi-periodic time scales. In difference to the previouswork (see above), the authors conclude that magnetic inter-action effectively locks the stellar rotation to about 10% thebreak-up velocity, a value which actually depends on thedisk accretion rate and the stellar magnetic moment. Al-though the authors correctly stress the importance of deal-ing with the exact balance between open and closed fieldlines and their corresponding angular momentum flux, theexact magnetospheric structure in the innermost region willcertainly depend on the resistivity (diffusivity) of the diskand corona material. This is, however, not included in theirtreatment as an ideal MHD code has been applied.

We summarize this section noting that tremendousprogress has occurred in the numerical simulation of thestar-disk interaction. The role of numerical simulationsis pivotal in this area because the mechanisms involvedare complex, strongly interrelated, and often highly time-dependent. It is fair to say that numerical simulations of thestar-disk interaction have not yet shown the launching of ajet flow comparable to the observations.

6. JETS AND GRAVITATIONAL COLLAPSE

Jets are expected to be associated with gravitational col-lapse because disks are the result of the collapse of rotatingmolecular cloud cores. One of the first simulations to showhow jets arise during gravitational collapse is the work ofTomisaka(1998, 2002). Here, the collapse of a magnetizedcore within a rotating cylinder of gas gave rise the formationof a disk from which a centrifugally drive, disk wind wasproduced. Analytical work byKrasnopolsky and K̈onigl(2002) provides an interesting class of self-similar solutionsfor the collapse of rotating, magnetized, isothermal coresincluding ambipolar diffusion that could be matched ontoBP82 outflows.

Given the importance of Bonner-Ebert (B-E) spheres inthe physics of star formation and gravitational collapse, re-cent efforts have focused on understanding the evolution ofmagnetized B-E spheres. Whereas purely hydrodynamiccollapses of such objects never show outflows (e.g.,Fos-ter and Chevalier, 1993;Banerjee et al., 2004), the addi-tion of a magnetic field produces them.Matsumoto andTomisaka(2004) have studied the collapse of rotating B-Espheres wherein the magnetic axis is inclined with the ini-tial rotation axis. They observe that after the formation ofanadiabatic core, outflow is ejected along the local magneticfield lines. Eventually, strong torques result in the rapidalignment of the rotation and magnetic field vectors.

The collapse of a magnetized, rotating, B-E sphere withmolecular cooling included, was carried out byBanerjeeand Pudritz(2006) using the FLASH adaptive mesh re-finement code. The results of this simulation are shownin Fig. 5 which shows the end state (at about 70,000 yrs)of the collapse of a B-E sphere that is chosen to be pre-cisely the Bok globule observed byAlves et al.(2001) –whose mass is2.1M� and radiusR = 1.25 × 104 AU atan initial temperature of 16 K. Two types of outflow can be

15

Fig. 5.— Large scale outflow (top frame, scale of hundredsof AU), and and small scale disk wind and jet formed (bot-tom frame, scale of a fraction of an AU) during the gravi-tational collapse of a magnetized B-E, rotating cloud core.Cross-sections through the disk and outflows are shown –the blue contour marks the Alfv́en surface. Snapshots takenof an Adaptive Mesh calculation at about 70,000 years intothe collapse. Adapted fromBanerjee and Pudritz(2006).

seen: (i) an outflow that originates at scale of' 130 AUon the forming disk that consists of a wound up column oftoroidal magnetic field whose pressure gradient pushes outa slow outflow; and (ii) a disk-wind that collimates into ajet on scale of 0.07 AU. A tight proto-binary system hasformed in this simulation, whose masses are still very small≤ 10−2M�, which is much less than the mass of the diskat this time' 10−1M�. The outer flow bears the hallmarkof a magnetic tower, first observed byUchida and Shibata,and studied byLynden-Bell(2003). Both flow componentsare unbound, with the disk wind reaching3 km s−1 at 0.4

AU which is quite super-Alfvénic and above the local es-cape speed. The outflow and jet speeds will increase as thecentral mass grows.

We conclude that the theory and computation of jets andoutflows is in excellent agreement with new observationsof many kinds. Disk winds are triggered during magne-tized collapse and persist throughout the evolution of thedisk. They efficiently tap accretion power, transport a sig-nificant part portion of the disk’s angular momentum, andcan achieve different degrees of collimation depending ontheir mass loading. Accretion-powered outflows may alsosolve the stellar angular momentum problem. We are opti-mistic that observations will soon be able to test the univer-sality of outflows, all the way from circumplanetary disksto those around O stars.

Acknowledgments. We are grateful to the organizersfor the opportunity to write this chapter, Nordita for hostingan authors’ meeting, Tom Ray for stimulating discussions,Robi Banerjee for a careful read of the manuscript, and ananonymous referee for a useful report. The research of REPand RO is supported by grants from NSERC of Canada.C.F. acknowledges travel support by the German sciencefoundation to participate in the PPV conference.

REFERENCES

Afshordi N., Mukhopadhyay B., and Narayan R. (2005)Astro-phys. J., 629, 373-382.

Alves J. F., Lada C. J., Lada E. A. (2001)Nature, 409, 159-161.Anderson J. M., Li Z.-Y., Krasnopolsky R., and Blandford R. D.

(2003)Astrophys. J., 590, L107-L110.Anderson J. M., Li Z.-Y., Krasnopolsky R., and Blandford R. D.

(2005)Astrophys. J., 630, 945-957.Bacciotti F., Mundt R., Ray T. P., Eislöffel J., Solf J., andCamen-

zind M. (2000)Astrophys. J., 537, L49-L52.Bacciotti F., Ray T.